Monitoring Collective Communication Among GPUs